Can Hessian-Based Insights Support Fault Diagnosis in Attention-based Models?
Not on the map. Why?
Abstract (arXiv)
As attention-based deep learning models scale in size and complexity, diagnosing their faults becomes increasingly challenging. In this work, we conduct an empirical study to evaluate the potential of Hessian-based analysis for diagnosing faults in attention-based models. Specifically, we use Hessian-derived insights to identify fragile regions (via curvature analysis) and parameter interdependencies (via parameter interaction analysis) within attention mechanisms. Through experiments on three diverse models (HAN, 3D-CNN, DistilBERT), we show that Hessian-based metrics can localize instability and pinpoint fault sources more effectively than gradients alone. Our empirical findings suggest that these metrics could significantly improve fault diagnosis in complex neural architectures, potentially improving software debugging practices.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 35 of its 54 references that are in the map. First posted on arXiv 9 Jun 2025.
Builds on
35 of 54 references.
References · 35
- 2017Attention Is All You Needcited by 60,361
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
- 2014Explaining and Harnessing Adversarial Examplescited by 7,364
- 2020Scaling Laws for Neural Language Modelscited by 6,586
- 2018Sanity Checks for Saliency Mapscited by 1,156
- 2017Visualizing the Loss Landscape of Neural Netscited by 1,130
- 2019Attention is not Explanationcited by 929
- 2021GPT Understands, Toocited by 716
- 2023Scaling Vision Transformers to 22 Billion Parameterscited by 694
- 2017Interpretation of Neural Networks is Fragilecited by 444
- 2021Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depthcited by 419
- 2019Is Attention Interpretable?cited by 389
23 more
- 2023Studying Large Language Model Generalization with Influence Functionscited by 326
- 2019An Investigation into Neural Net Optimization via Hessian Eigenvalue Densitycited by 292
- 2019PyHessian: Neural Networks Through the Lens of the Hessiancited by 292
- 2014Identifying and attacking the saddle point problem in high-dimensional non-convex optimizationcited by 289
- 2019Understanding and Improving Layer Normalizationcited by 217
- 2014Qualitatively characterizing neural network optimization problemscited by 188
- 2023Small-scale proxies for large-scale Transformer training instabilitiescited by 183
- 2020The Break-Even Point on Optimization Trajectories of Deep Neural Networkscited by 150
- 2017Deep Learning based Recommender System: A Survey and New Perspectivescited by 136
- 2018Hessian-based Analysis of Large Batch Training and Robustness to Adversariescited by 87
- 2021Analytic Insights into Structure and Rank of Neural Network Hessian Mapscited by 52
- 2024Dynamic metastability in the self-attention modelcited by 46
- 2020Why Attentions May Not Be Interpretable?cited by 41
- 2020A Survey on Deep Learning for Software Engineeringcited by 38
- 2021DeepLocalize: Fault Localization for Deep Neural Networkscited by 38
- 2019On Model Stability as a Function of Random Seedcited by 31
- 2022Stability and Generalization of Stochastic Optimization with Nonconvex and Nonsmooth Problemscited by 25
- 2019Negative eigenvalues of the Hessian in deep neural networkscited by 20
- 2020Piecewise linear activations substantially shape the loss surfaces of neural networkscited by 19
- 2022Efficiently Training Low-Curvature Neural Networkscited by 19
- 2022DeepFD: Automated Fault Diagnosis and Localization for Deep Learning Programscited by 16
- 2023Effects of noise on the overparametrization of quantum neural networkscited by 9
- 2023No Wrong Turns: The Simple Geometry Of Neural Networks Optimization Pathscited by 7
Built on by
Citing papers per year: 2025: 1; 2026: 1.
2 papers.